LOVE/love
1
0
Fork 0
mirror of https://github.com/love2d/love.git synced 2026-08-18 19:54:22 +02:00
Code Issues Packages Projects Releases Wiki Activity

Update spirv-cross

Browse Source
This commit is contained in:
Alex Szpakowski
2020-06-06 16:42:12 -03:00
parent de041ded0e
commit f5ecaf2e92
16 changed files with 845 additions and 265 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/libraries/spirv_cross/spirv_hlsl.cpp
+386 -123
View File
@@ -23,6 +23,41 @@ using namespace spv;
using namespace SPIRV_CROSS_NAMESPACE;
using namespace std;
enum class ImageFormatNormalizedState
{
None = 0,
Unorm = 1,
Snorm = 2
};
static ImageFormatNormalizedState image_format_to_normalized_state(ImageFormat fmt)
{
switch (fmt)
{
case ImageFormatR8:
case ImageFormatR16:
case ImageFormatRg8:
case ImageFormatRg16:
case ImageFormatRgba8:
case ImageFormatRgba16:
case ImageFormatRgb10A2:
return ImageFormatNormalizedState::Unorm;
case ImageFormatR8Snorm:
case ImageFormatR16Snorm:
case ImageFormatRg8Snorm:
case ImageFormatRg16Snorm:
case ImageFormatRgba8Snorm:
case ImageFormatRgba16Snorm:
return ImageFormatNormalizedState::Snorm;
default:
break;
}
return ImageFormatNormalizedState::None;
}
static unsigned image_format_to_components(ImageFormat fmt)
{
switch (fmt)
@@ -395,7 +430,20 @@ string CompilerHLSL::type_to_glsl(const SPIRType &type, uint32_t id)
case SPIRType::AtomicCounter:
return "atomic_uint";
case SPIRType::Half:
return "min16float";
if (hlsl_options.enable_16bit_types)
return "half";
else
return "min16float";
case SPIRType::Short:
if (hlsl_options.enable_16bit_types)
return "int16_t";
else
return "min16int";
case SPIRType::UShort:
if (hlsl_options.enable_16bit_types)
return "uint16_t";
else
return "min16uint";
case SPIRType::Float:
return "float";
case SPIRType::Double:
@@ -423,7 +471,11 @@ string CompilerHLSL::type_to_glsl(const SPIRType &type, uint32_t id)
case SPIRType::UInt:
return join("uint", type.vecsize);
case SPIRType::Half:
return join("min16float", type.vecsize);
return join(hlsl_options.enable_16bit_types ? "half" : "min16float", type.vecsize);
case SPIRType::Short:
return join(hlsl_options.enable_16bit_types ? "int16_t" : "min16int", type.vecsize);
case SPIRType::UShort:
return join(hlsl_options.enable_16bit_types ? "uint16_t" : "min16uint", type.vecsize);
case SPIRType::Float:
return join("float", type.vecsize);
case SPIRType::Double:
@@ -447,7 +499,11 @@ string CompilerHLSL::type_to_glsl(const SPIRType &type, uint32_t id)
case SPIRType::UInt:
return join("uint", type.columns, "x", type.vecsize);
case SPIRType::Half:
return join("min16float", type.columns, "x", type.vecsize);
return join(hlsl_options.enable_16bit_types ? "half" : "min16float", type.columns, "x", type.vecsize);
case SPIRType::Short:
return join(hlsl_options.enable_16bit_types ? "int16_t" : "min16int", type.columns, "x", type.vecsize);
case SPIRType::UShort:
return join(hlsl_options.enable_16bit_types ? "uint16_t" : "min16uint", type.columns, "x", type.vecsize);
case SPIRType::Float:
return join("float", type.columns, "x", type.vecsize);
case SPIRType::Double:
@@ -1423,66 +1479,14 @@ void CompilerHLSL::emit_resources()
}
}
if (required_textureSizeVariants != 0)
emit_texture_size_variants(required_texture_size_variants.srv, "4", false, "");
for (uint32_t norm = 0; norm < 3; norm++)
{
static const char *types[QueryTypeCount] = { "float4", "int4", "uint4" };
static const char *dims[QueryDimCount] = { "Texture1D", "Texture1DArray", "Texture2D", "Texture2DArray",
"Texture3D", "Buffer", "TextureCube", "TextureCubeArray",
"Texture2DMS", "Texture2DMSArray" };
static const bool has_lod[QueryDimCount] = { true, true, true, true, true, false, true, true, false, false };
static const char *ret_types[QueryDimCount] = {
"uint", "uint2", "uint2", "uint3", "uint3", "uint", "uint2", "uint3", "uint2", "uint3",
};
static const uint32_t return_arguments[QueryDimCount] = {
1, 2, 2, 3, 3, 1, 2, 3, 2, 3,
};
for (uint32_t index = 0; index < QueryDimCount; index++)
for (uint32_t comp = 0; comp < 4; comp++)
{
for (uint32_t type_index = 0; type_index < QueryTypeCount; type_index++)
{
uint32_t bit = 16 * type_index + index;
uint64_t mask = 1ull << bit;
if ((required_textureSizeVariants & mask) == 0)
continue;
statement(ret_types[index], " SPIRV_Cross_textureSize(", dims[index], "<", types[type_index],
"> Tex, uint Level, out uint Param)");
begin_scope();
statement(ret_types[index], " ret;");
switch (return_arguments[index])
{
case 1:
if (has_lod[index])
statement("Tex.GetDimensions(Level, ret.x, Param);");
else
{
statement("Tex.GetDimensions(ret.x);");
statement("Param = 0u;");
}
break;
case 2:
if (has_lod[index])
statement("Tex.GetDimensions(Level, ret.x, ret.y, Param);");
else
statement("Tex.GetDimensions(ret.x, ret.y, Param);");
break;
case 3:
if (has_lod[index])
statement("Tex.GetDimensions(Level, ret.x, ret.y, ret.z, Param);");
else
statement("Tex.GetDimensions(ret.x, ret.y, ret.z, Param);");
break;
}
statement("return ret;");
end_scope();
statement("");
}
static const char *qualifiers[] = { "", "unorm ", "snorm " };
static const char *vecsizes[] = { "", "2", "3", "4" };
emit_texture_size_variants(required_texture_size_variants.uav[norm][comp], vecsizes[comp], true, qualifiers[norm]);
}
}
@@ -1845,6 +1849,83 @@ void CompilerHLSL::emit_resources()
}
}
void CompilerHLSL::emit_texture_size_variants(uint64_t variant_mask, const char *vecsize_qualifier, bool uav, const char *type_qualifier)
{
if (variant_mask == 0)
return;
static const char *types[QueryTypeCount] = { "float", "int", "uint" };
static const char *dims[QueryDimCount] = { "Texture1D", "Texture1DArray", "Texture2D", "Texture2DArray",
"Texture3D", "Buffer", "TextureCube", "TextureCubeArray",
"Texture2DMS", "Texture2DMSArray" };
static const bool has_lod[QueryDimCount] = { true, true, true, true, true, false, true, true, false, false };
static const char *ret_types[QueryDimCount] = {
"uint", "uint2", "uint2", "uint3", "uint3", "uint", "uint2", "uint3", "uint2", "uint3",
};
static const uint32_t return_arguments[QueryDimCount] = {
1, 2, 2, 3, 3, 1, 2, 3, 2, 3,
};
for (uint32_t index = 0; index < QueryDimCount; index++)
{
for (uint32_t type_index = 0; type_index < QueryTypeCount; type_index++)
{
uint32_t bit = 16 * type_index + index;
uint64_t mask = 1ull << bit;
if ((variant_mask & mask) == 0)
continue;
statement(ret_types[index], " SPIRV_Cross_", (uav ? "image" : "texture"), "Size(", (uav ? "RW" : ""),
dims[index], "<", type_qualifier, types[type_index], vecsize_qualifier,
"> Tex, ", (uav ? "" : "uint Level, "), "out uint Param)");
begin_scope();
statement(ret_types[index], " ret;");
switch (return_arguments[index])
{
case 1:
if (has_lod[index] && !uav)
statement("Tex.GetDimensions(Level, ret.x, Param);");
else
{
statement("Tex.GetDimensions(ret.x);");
statement("Param = 0u;");
}
break;
case 2:
if (has_lod[index] && !uav)
statement("Tex.GetDimensions(Level, ret.x, ret.y, Param);");
else if (!uav)
statement("Tex.GetDimensions(ret.x, ret.y, Param);");
else
{
statement("Tex.GetDimensions(ret.x, ret.y);");
statement("Param = 0u;");
}
break;
case 3:
if (has_lod[index] && !uav)
statement("Tex.GetDimensions(Level, ret.x, ret.y, ret.z, Param);");
else if (!uav)
statement("Tex.GetDimensions(ret.x, ret.y, ret.z, Param);");
else
{
statement("Tex.GetDimensions(ret.x, ret.y, ret.z);");
statement("Param = 0u;");
}
break;
}
statement("return ret;");
end_scope();
statement("");
}
}
}
string CompilerHLSL::layout_for_member(const SPIRType &type, uint32_t index)
{
auto &flags = get_member_decoration_bitset(type.self, index);
@@ -1906,7 +1987,7 @@ void CompilerHLSL::emit_buffer_block(const SPIRVariable &var)
if (is_uav)
{
Bitset flags = ir.get_buffer_block_flags(var);
bool is_readonly = flags.get(DecorationNonWritable) && !hlsl_options.force_storage_buffer_as_uav;
bool is_readonly = flags.get(DecorationNonWritable) && !is_hlsl_force_storage_buffer_as_uav(var.self);
bool is_coherent = flags.get(DecorationCoherent) && !is_readonly;
bool is_interlocked = interlocked_resources.count(var.self) > 0;
const char *type_name = "ByteAddressBuffer ";
@@ -3038,7 +3119,7 @@ string CompilerHLSL::to_resource_binding(const SPIRVariable &var)
if (has_decoration(type.self, DecorationBufferBlock))
{
Bitset flags = ir.get_buffer_block_flags(var);
bool is_readonly = flags.get(DecorationNonWritable) && !hlsl_options.force_storage_buffer_as_uav;
bool is_readonly = flags.get(DecorationNonWritable) && !is_hlsl_force_storage_buffer_as_uav(var.self);
space = is_readonly ? 't' : 'u'; // UAV
resource_flags = is_readonly ? HLSL_BINDING_AUTO_SRV_BIT : HLSL_BINDING_AUTO_UAV_BIT;
}
@@ -3057,7 +3138,7 @@ string CompilerHLSL::to_resource_binding(const SPIRVariable &var)
{
// UAV or SRV depending on readonly flag.
Bitset flags = ir.get_buffer_block_flags(var);
bool is_readonly = flags.get(DecorationNonWritable) && !hlsl_options.force_storage_buffer_as_uav;
bool is_readonly = flags.get(DecorationNonWritable) && !is_hlsl_force_storage_buffer_as_uav(var.self);
space = is_readonly ? 't' : 'u';
resource_flags = is_readonly ? HLSL_BINDING_AUTO_SRV_BIT : HLSL_BINDING_AUTO_UAV_BIT;
}
@@ -3587,11 +3668,16 @@ void CompilerHLSL::read_access_chain(string *expr, const string &lhs, const SPIR
read_access_chain_struct(lhs, chain);
return;
}
else if (type.width != 32)
SPIRV_CROSS_THROW("Reading types other than 32-bit from ByteAddressBuffer not yet supported.");
else if (type.width != 32 && !hlsl_options.enable_16bit_types)
SPIRV_CROSS_THROW("Reading types other than 32-bit from ByteAddressBuffer not yet supported, unless SM 6.2 and native 16-bit types are enabled.");
bool templated_load = hlsl_options.shader_model >= 62;
string load_expr;
string template_expr;
if (templated_load)
template_expr = join("<", type_to_glsl(type), ">");
// Load a vector or scalar.
if (type.columns == 1 && !chain.row_major_matrix)
{
@@ -3614,12 +3700,24 @@ void CompilerHLSL::read_access_chain(string *expr, const string &lhs, const SPIR
SPIRV_CROSS_THROW("Unknown vector size.");
}
load_expr = join(chain.base, ".", load_op, "(", chain.dynamic_index, chain.static_index, ")");
if (templated_load)
load_op = "Load";
load_expr = join(chain.base, ".", load_op, template_expr, "(", chain.dynamic_index, chain.static_index, ")");
}
else if (type.columns == 1)
{
// Strided load since we are loading a column from a row-major matrix.
if (type.vecsize > 1)
if (templated_load)
{
auto scalar_type = type;
scalar_type.vecsize = 1;
scalar_type.columns = 1;
template_expr = join("<", type_to_glsl(scalar_type), ">");
if (type.vecsize > 1)
load_expr += type_to_glsl(type) + "(";
}
else if (type.vecsize > 1)
{
load_expr = type_to_glsl(target_type);
load_expr += "(";
@@ -3628,7 +3726,7 @@ void CompilerHLSL::read_access_chain(string *expr, const string &lhs, const SPIR
for (uint32_t r = 0; r < type.vecsize; r++)
{
load_expr +=
join(chain.base, ".Load(", chain.dynamic_index, chain.static_index + r * chain.matrix_stride, ")");
join(chain.base, ".Load", template_expr, "(", chain.dynamic_index, chain.static_index + r * chain.matrix_stride, ")");
if (r + 1 < type.vecsize)
load_expr += ", ";
}
@@ -3658,13 +3756,25 @@ void CompilerHLSL::read_access_chain(string *expr, const string &lhs, const SPIR
SPIRV_CROSS_THROW("Unknown vector size.");
}
// Note, this loading style in HLSL is *actually* row-major, but we always treat matrices as transposed in this backend,
// so row-major is technically column-major ...
load_expr = type_to_glsl(target_type);
if (templated_load)
{
auto vector_type = type;
vector_type.columns = 1;
template_expr = join("<", type_to_glsl(vector_type), ">");
load_expr = type_to_glsl(type);
load_op = "Load";
}
else
{
// Note, this loading style in HLSL is *actually* row-major, but we always treat matrices as transposed in this backend,
// so row-major is technically column-major ...
load_expr = type_to_glsl(target_type);
}
load_expr += "(";
for (uint32_t c = 0; c < type.columns; c++)
{
load_expr += join(chain.base, ".", load_op, "(", chain.dynamic_index,
load_expr += join(chain.base, ".", load_op, template_expr, "(", chain.dynamic_index,
chain.static_index + c * chain.matrix_stride, ")");
if (c + 1 < type.columns)
load_expr += ", ";
@@ -3676,13 +3786,24 @@ void CompilerHLSL::read_access_chain(string *expr, const string &lhs, const SPIR
// Pick out elements one by one ... Hopefully compilers are smart enough to recognize this pattern
// considering HLSL is "row-major decl", but "column-major" memory layout (basically implicit transpose model, ugh) ...
load_expr = type_to_glsl(target_type);
if (templated_load)
{
load_expr = type_to_glsl(type);
auto scalar_type = type;
scalar_type.vecsize = 1;
scalar_type.columns = 1;
template_expr = join("<", type_to_glsl(scalar_type), ">");
}
else
load_expr = type_to_glsl(target_type);
load_expr += "(";
for (uint32_t c = 0; c < type.columns; c++)
{
for (uint32_t r = 0; r < type.vecsize; r++)
{
load_expr += join(chain.base, ".Load(", chain.dynamic_index,
load_expr += join(chain.base, ".Load", template_expr, "(", chain.dynamic_index,
chain.static_index + c * (type.width / 8) + r * chain.matrix_stride, ")");
if ((r + 1 < type.vecsize) || (c + 1 < type.columns))
@@ -3692,9 +3813,12 @@ void CompilerHLSL::read_access_chain(string *expr, const string &lhs, const SPIR
load_expr += ")";
}
auto bitcast_op = bitcast_glsl_op(type, target_type);
if (!bitcast_op.empty())
load_expr = join(bitcast_op, "(", load_expr, ")");
if (!templated_load)
{
auto bitcast_op = bitcast_glsl_op(type, target_type);
if (!bitcast_op.empty())
load_expr = join(bitcast_op, "(", load_expr, ")");
}
if (lhs.empty())
{
@@ -3877,8 +4001,14 @@ void CompilerHLSL::write_access_chain(const SPIRAccessChain &chain, uint32_t val
register_write(chain.self);
return;
}
else if (type.width != 32)
SPIRV_CROSS_THROW("Writing types other than 32-bit to RWByteAddressBuffer not yet supported.");
else if (type.width != 32 && !hlsl_options.enable_16bit_types)
SPIRV_CROSS_THROW("Writing types other than 32-bit to RWByteAddressBuffer not yet supported, unless SM 6.2 and native 16-bit types are enabled.");
bool templated_store = hlsl_options.shader_model >= 62;
string template_expr;
if (templated_store)
template_expr = join("<", type_to_glsl(type), ">");
if (type.columns == 1 && !chain.row_major_matrix)
{
@@ -3902,13 +4032,27 @@ void CompilerHLSL::write_access_chain(const SPIRAccessChain &chain, uint32_t val
}
auto store_expr = write_access_chain_value(value, composite_chain, false);
auto bitcast_op = bitcast_glsl_op(target_type, type);
if (!bitcast_op.empty())
store_expr = join(bitcast_op, "(", store_expr, ")");
statement(chain.base, ".", store_op, "(", chain.dynamic_index, chain.static_index, ", ", store_expr, ");");
if (!templated_store)
{
auto bitcast_op = bitcast_glsl_op(target_type, type);
if (!bitcast_op.empty())
store_expr = join(bitcast_op, "(", store_expr, ")");
}
else
store_op = "Store";
statement(chain.base, ".", store_op, template_expr, "(", chain.dynamic_index, chain.static_index, ", ", store_expr, ");");
}
else if (type.columns == 1)
{
if (templated_store)
{
auto scalar_type = type;
scalar_type.vecsize = 1;
scalar_type.columns = 1;
template_expr = join("<", type_to_glsl(scalar_type), ">");
}
// Strided store.
for (uint32_t r = 0; r < type.vecsize; r++)
{
@@ -3920,10 +4064,14 @@ void CompilerHLSL::write_access_chain(const SPIRAccessChain &chain, uint32_t val
}
remove_duplicate_swizzle(store_expr);
auto bitcast_op = bitcast_glsl_op(target_type, type);
if (!bitcast_op.empty())
store_expr = join(bitcast_op, "(", store_expr, ")");
statement(chain.base, ".Store(", chain.dynamic_index, chain.static_index + chain.matrix_stride * r, ", ",
if (!templated_store)
{
auto bitcast_op = bitcast_glsl_op(target_type, type);
if (!bitcast_op.empty())
store_expr = join(bitcast_op, "(", store_expr, ")");
}
statement(chain.base, ".Store", template_expr, "(", chain.dynamic_index, chain.static_index + chain.matrix_stride * r, ", ",
store_expr, ");");
}
}
@@ -3948,18 +4096,39 @@ void CompilerHLSL::write_access_chain(const SPIRAccessChain &chain, uint32_t val
SPIRV_CROSS_THROW("Unknown vector size.");
}
if (templated_store)
{
store_op = "Store";
auto vector_type = type;
vector_type.columns = 1;
template_expr = join("<", type_to_glsl(vector_type), ">");
}
for (uint32_t c = 0; c < type.columns; c++)
{
auto store_expr = join(write_access_chain_value(value, composite_chain, true), "[", c, "]");
auto bitcast_op = bitcast_glsl_op(target_type, type);
if (!bitcast_op.empty())
store_expr = join(bitcast_op, "(", store_expr, ")");
statement(chain.base, ".", store_op, "(", chain.dynamic_index, chain.static_index + c * chain.matrix_stride,
if (!templated_store)
{
auto bitcast_op = bitcast_glsl_op(target_type, type);
if (!bitcast_op.empty())
store_expr = join(bitcast_op, "(", store_expr, ")");
}
statement(chain.base, ".", store_op, template_expr, "(", chain.dynamic_index, chain.static_index + c * chain.matrix_stride,
", ", store_expr, ");");
}
}
else
{
if (templated_store)
{
auto scalar_type = type;
scalar_type.vecsize = 1;
scalar_type.columns = 1;
template_expr = join("<", type_to_glsl(scalar_type), ">");
}
for (uint32_t r = 0; r < type.vecsize; r++)
{
for (uint32_t c = 0; c < type.columns; c++)
@@ -3970,7 +4139,7 @@ void CompilerHLSL::write_access_chain(const SPIRAccessChain &chain, uint32_t val
auto bitcast_op = bitcast_glsl_op(target_type, type);
if (!bitcast_op.empty())
store_expr = join(bitcast_op, "(", store_expr, ")");
statement(chain.base, ".Store(", chain.dynamic_index,
statement(chain.base, ".Store", template_expr, "(", chain.dynamic_index,
chain.static_index + c * (type.width / 8) + r * chain.matrix_stride, ", ", store_expr, ");");
}
}
@@ -4087,9 +4256,11 @@ void CompilerHLSL::emit_atomic(const uint32_t *ops, uint32_t length, spv::Op op)
const char *atomic_op = nullptr;
string value_expr;
if (op != OpAtomicIDecrement && op != OpAtomicIIncrement)
if (op != OpAtomicIDecrement && op != OpAtomicIIncrement && op != OpAtomicLoad && op != OpAtomicStore)
value_expr = to_expression(ops[op == OpAtomicCompareExchange ? 6 : 5]);
bool is_atomic_store = false;
switch (op)
{
case OpAtomicIIncrement:
@@ -4102,6 +4273,11 @@ void CompilerHLSL::emit_atomic(const uint32_t *ops, uint32_t length, spv::Op op)
value_expr = "-1";
break;
case OpAtomicLoad:
atomic_op = "InterlockedAdd";
value_expr = "0";
break;
case OpAtomicISub:
atomic_op = "InterlockedAdd";
value_expr = join("-", enclose_expression(value_expr));
@@ -4137,6 +4313,11 @@ void CompilerHLSL::emit_atomic(const uint32_t *ops, uint32_t length, spv::Op op)
atomic_op = "InterlockedExchange";
break;
case OpAtomicStore:
atomic_op = "InterlockedExchange";
is_atomic_store = true;
break;
case OpAtomicCompareExchange:
if (length < 8)
SPIRV_CROSS_THROW("Not enough data for opcode.");
@@ -4148,31 +4329,57 @@ void CompilerHLSL::emit_atomic(const uint32_t *ops, uint32_t length, spv::Op op)
SPIRV_CROSS_THROW("Unknown atomic opcode.");
}
uint32_t result_type = ops[0];
uint32_t id = ops[1];
forced_temporaries.insert(ops[1]);
auto &type = get<SPIRType>(result_type);
statement(variable_decl(type, to_name(id)), ";");
auto &data_type = expression_type(ops[2]);
auto *chain = maybe_get<SPIRAccessChain>(ops[2]);
SPIRType::BaseType expr_type;
if (data_type.storage == StorageClassImage || !chain)
if (is_atomic_store)
{
statement(atomic_op, "(", to_expression(ops[2]), ", ", value_expr, ", ", to_name(id), ");");
expr_type = data_type.basetype;
auto &data_type = expression_type(ops[0]);
auto *chain = maybe_get<SPIRAccessChain>(ops[0]);
auto &tmp_id = extra_sub_expressions[ops[0]];
if (!tmp_id)
{
tmp_id = ir.increase_bound_by(1);
emit_uninitialized_temporary_expression(get_pointee_type(data_type).self, tmp_id);
}
if (data_type.storage == StorageClassImage || !chain)
{
statement(atomic_op, "(", to_expression(ops[0]), ", ", to_expression(ops[3]), ", ", to_expression(tmp_id), ");");
}
else
{
// RWByteAddress buffer is always uint in its underlying type.
statement(chain->base, ".", atomic_op, "(", chain->dynamic_index, chain->static_index, ", ", to_expression(ops[3]),
", ", to_expression(tmp_id), ");");
}
}
else
{
// RWByteAddress buffer is always uint in its underlying type.
expr_type = SPIRType::UInt;
statement(chain->base, ".", atomic_op, "(", chain->dynamic_index, chain->static_index, ", ", value_expr, ", ",
to_name(id), ");");
}
uint32_t result_type = ops[0];
uint32_t id = ops[1];
forced_temporaries.insert(ops[1]);
auto expr = bitcast_expression(type, expr_type, to_name(id));
set<SPIRExpression>(id, expr, result_type, true);
auto &type = get<SPIRType>(result_type);
statement(variable_decl(type, to_name(id)), ";");
auto &data_type = expression_type(ops[2]);
auto *chain = maybe_get<SPIRAccessChain>(ops[2]);
SPIRType::BaseType expr_type;
if (data_type.storage == StorageClassImage || !chain)
{
statement(atomic_op, "(", to_expression(ops[2]), ", ", value_expr, ", ", to_name(id), ");");
expr_type = data_type.basetype;
}
else
{
// RWByteAddress buffer is always uint in its underlying type.
expr_type = SPIRType::UInt;
statement(chain->base, ".", atomic_op, "(", chain->dynamic_index, chain->static_index, ", ", value_expr,
", ", to_name(id), ");");
}
auto expr = bitcast_expression(type, expr_type, to_name(id));
set<SPIRExpression>(id, expr, result_type, true);
}
flush_all_atomic_capable_variables();
}
@@ -4796,8 +5003,7 @@ void CompilerHLSL::emit_instruction(const Instruction &instruction)
auto result_type = ops[0];
auto id = ops[1];
require_texture_query_variant(expression_type(ops[2]));
require_texture_query_variant(ops[2]);
auto dummy_samples_levels = join(get_fallback_name(id), "_dummy_parameter");
statement("uint ", dummy_samples_levels, ";");
@@ -4815,12 +5021,22 @@ void CompilerHLSL::emit_instruction(const Instruction &instruction)
auto result_type = ops[0];
auto id = ops[1];
require_texture_query_variant(expression_type(ops[2]));
require_texture_query_variant(ops[2]);
bool uav = expression_type(ops[2]).image.sampled == 2;
if (const auto *var = maybe_get_backing_variable(ops[2]))
if (hlsl_options.nonwritable_uav_texture_as_srv && has_decoration(var->self, DecorationNonWritable))
uav = false;
auto dummy_samples_levels = join(get_fallback_name(id), "_dummy_parameter");
statement("uint ", dummy_samples_levels, ";");
auto expr = join("SPIRV_Cross_textureSize(", to_expression(ops[2]), ", 0u, ", dummy_samples_levels, ")");
string expr;
if (uav)
expr = join("SPIRV_Cross_imageSize(", to_expression(ops[2]), ", ", dummy_samples_levels, ")");
else
expr = join("SPIRV_Cross_textureSize(", to_expression(ops[2]), ", 0u, ", dummy_samples_levels, ")");
auto &restype = get<SPIRType>(ops[0]);
expr = bitcast_expression(restype, SPIRType::UInt, expr);
emit_op(result_type, id, expr, true);
@@ -4833,14 +5049,25 @@ void CompilerHLSL::emit_instruction(const Instruction &instruction)
auto result_type = ops[0];
auto id = ops[1];
require_texture_query_variant(expression_type(ops[2]));
require_texture_query_variant(ops[2]);
bool uav = expression_type(ops[2]).image.sampled == 2;
if (opcode == OpImageQueryLevels && uav)
SPIRV_CROSS_THROW("Cannot query levels for UAV images.");
if (const auto *var = maybe_get_backing_variable(ops[2]))
if (hlsl_options.nonwritable_uav_texture_as_srv && has_decoration(var->self, DecorationNonWritable))
uav = false;
// Keep it simple and do not emit special variants to make this look nicer ...
// This stuff is barely, if ever, used.
forced_temporaries.insert(id);
auto &type = get<SPIRType>(result_type);
statement(variable_decl(type, to_name(id)), ";");
statement("SPIRV_Cross_textureSize(", to_expression(ops[2]), ", 0u, ", to_name(id), ");");
if (uav)
statement("SPIRV_Cross_imageSize(", to_expression(ops[2]), ", ", to_name(id), ");");
else
statement("SPIRV_Cross_textureSize(", to_expression(ops[2]), ", 0u, ", to_name(id), ");");
auto &restype = get<SPIRType>(ops[0]);
auto expr = bitcast_expression(restype, SPIRType::UInt, to_name(id));
@@ -4967,6 +5194,8 @@ void CompilerHLSL::emit_instruction(const Instruction &instruction)
case OpAtomicIAdd:
case OpAtomicIIncrement:
case OpAtomicIDecrement:
case OpAtomicLoad:
case OpAtomicStore:
{
emit_atomic(ops, instruction.length, opcode);
break;
@@ -5152,8 +5381,16 @@ void CompilerHLSL::emit_instruction(const Instruction &instruction)
}
}
void CompilerHLSL::require_texture_query_variant(const SPIRType &type)
void CompilerHLSL::require_texture_query_variant(uint32_t var_id)
{
if (const auto *var = maybe_get_backing_variable(var_id))
var_id = var->self;
auto &type = expression_type(var_id);
bool uav = type.image.sampled == 2;
if (hlsl_options.nonwritable_uav_texture_as_srv && has_decoration(var_id, DecorationNonWritable))
uav = false;
uint32_t bit = 0;
switch (type.image.dim)
{
@@ -5202,11 +5439,15 @@ void CompilerHLSL::require_texture_query_variant(const SPIRType &type)
SPIRV_CROSS_THROW("Unsupported query type.");
}
auto norm_state = image_format_to_normalized_state(type.image.format);
auto &variant = uav ? required_texture_size_variants.uav[uint32_t(norm_state)][image_format_to_components(type.image.format) - 1] :
required_texture_size_variants.srv;
uint64_t mask = 1ull << bit;
if ((required_textureSizeVariants & mask) == 0)
if ((variant & mask) == 0)
{
force_recompile();
required_textureSizeVariants |= mask;
variant |= mask;
}
}
@@ -5291,6 +5532,9 @@ void CompilerHLSL::validate_shader_model()
if (ir.addressing_model != AddressingModelLogical)
SPIRV_CROSS_THROW("Only Logical addressing model can be used with HLSL.");
if (hlsl_options.enable_16bit_types && hlsl_options.shader_model < 62)
SPIRV_CROSS_THROW("Need at least shader model 6.2 when enabling native 16-bit type support.");
}
string CompilerHLSL::compile()
@@ -5413,3 +5657,22 @@ CompilerHLSL::BitcastType CompilerHLSL::get_bitcast_type(uint32_t result_type, u
return BitcastType::TypeNormal;
}
bool CompilerHLSL::is_hlsl_force_storage_buffer_as_uav(ID id) const
{
if (hlsl_options.force_storage_buffer_as_uav)
{
return true;
}
const uint32_t desc_set = get_decoration(id, spv::DecorationDescriptorSet);
const uint32_t binding = get_decoration(id, spv::DecorationBinding);
return (force_uav_buffer_bindings.find({desc_set, binding}) != force_uav_buffer_bindings.end());
}
void CompilerHLSL::set_hlsl_force_storage_buffer_as_uav(uint32_t desc_set, uint32_t binding)
{
SetBindingPair pair = { desc_set, binding };
force_uav_buffer_bindings.insert(pair);
}